requireAccess threw BadRequestException which is incorrect HTTP semantics.
Access denial is a client authorization problem (403 Forbidden), not a
malformed request (400 Bad Request). Keep the descriptive permission name
in the message since the full permission set is public API surface.
Album owners could probe arbitrary UUIDs via the add-user endpoint and
determine whether they belonged to registered accounts by receiving
'User not found'. The message is now ambiguous about whether the ID was
unrecognised or the user is inactive.
GET /users/:id/profile-image returned HTTP 400 for an unknown user ID
but HTTP 404 when the user existed without a photo, letting callers
distinguish the two cases. Both now return 404 so the response is
identical regardless of whether the UUID maps to an account.
Surfacing the Postgres unique-constraint name in the error response let
any authenticated user brute-force whether a custom slug was already in
use by another user's shared link, leaking the existence of other links.
Error messages on registration and profile-update that named whether an
email address was already taken allowed callers to enumerate registered
accounts. All three sites now return the same generic message regardless
of whether the address is in use.
Two distinct error messages in the OAuth callback endpoint revealed
whether an email address was already registered in the database.
An attacker controlling the OAuth provider's email claim could probe
the user table without authentication. Both cases now return the same
generic message.
Always call compareBcrypt in the login path regardless of whether the
email is registered. When no user is found, a dummy hash is used so the
bcrypt KDF still runs and response latency is constant, making it
impossible to enumerate valid email addresses by measuring response time.
* fix(ml): handle empty/corrupt images in face detection
When a corrupt or degenerate image with zero-dimension (0 width or 0 height)
reaches the face detection pipeline, insightface's RetinaFace.detect() calls
cv2.resize() with a target size of 0, triggering an OpenCV assertion failure:
error: (-215:Assertion failed) inv_scale_x > 0 in function 'resize'
This crashes the ML worker and returns a 500 error to the server.
Add an early return in FaceDetector._predict() that checks for zero-dimension
images after decoding and returns empty detection results instead of passing
them to the insightface model.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix(ml): move empty image validation to request level
Per review feedback, validate image dimensions in the predict endpoint
(returning 400) rather than in each model's _predict method. This
catches all zero-dimension images before they reach any model task.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix(ml): resolve mypy strict type error in predict endpoint
Use intermediate `decoded` variable so mypy knows `.width` and `.height`
are accessed on `Image`, not on `Image | str`.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
timonrieger